Quality of intervention delivery in a cluster randomised controlled trial: A qualitative observational study with lessons for fidelity

Abstract

Background: Understanding intervention fidelity is an essential part of the evaluation of complex interventions because fidelity not only affects the validity of trial findings, but also because studies of fidelity can be used to identify barriers and facilitators to successful implementation, and so provide important information about factors likely to impact the uptake of the intervention into clinical practice. Participant observation methods have been identified as being particularly valuable in studies of fidelity, yet are rarely used. This study aimed to use these methods to explore the quality of implementation of a complex intervention (Safewards) on mental health wards during a cluster randomised controlled trial. Specific aims were firstly to describe the different ways in which the intervention was implemented, and secondly to explore the contextual factors moderating the quality of intervention delivery, in order to inform ‘real world’ implementation of the intervention. Methods: Safewards was implemented on 16 mental health wards in England. We used Research Assistants (RAs) trained in participant observation to record qualitative observational data on the quality of intervention delivery (n = 565 observations). At the end of the trial, two focus groups were conducted with RAs, which were used to develop the coding framework. Data were analysed using thematic analysis. Results: There was substantial variation in intervention delivery between wards. We observed modifications to the intervention which were both fidelity consistent and inconsistent, and could enhance or dilute the intervention effects. We used these data to develop a typology which describes the different ways in which the intervention was delivered. This typology could be used as a tool to collect qualitative observational data about fidelity during trials. Moderators of Safewards implementation included systemic, interpersonal, and individual factors and patient responses to the intervention. Conclusions: Our study demonstrates how, with appropriate training in participant observation, RAs can collect highquality observational data about the quality of intervention delivery during a trial, giving a more complete picture of ‘fidelity’ than measurements of adherence alone.